Slashdot Mirror
Can Foursquare Data Predict Where You Live?
chicksdaddy writes "File this one under 'proof of the obvious,' but researchers at the recent 4th International Workshop on Location Based Social Networks presented a paper proving that your activity on Foursquare can be used to reliably determine your hometown. A study of data on 13 million Foursquare accounts showed that researchers could infer 'with high accuracy' where a particular user lives based on their accumulation of mayorships, check-ins and tips. Specifically: the researchers could correctly infer the home town of the Foursquare users 78% of the time, within an accuracy of about 50 kilometers."
That is tragically horrible accuracy. I was hoping the punchline would have been "to within 1 city block", 50km is comical.
This study funded by the Foundation for Obvious Studies, and will soon be published in the Journal for Obvious and Tautological Results.
In a follow up study, they'll figure out where you work, too.
That's pretty sad considering that close to 90% (citation needed) of foursquare users are the mayor of their own house.
Just within 50 km? You'd think that would be closer to 5k Hell, closer to 5 feet since I know many that are the mayor of their own house.
Possibly, depending on how often you post to Foursquare.
sysadmins and parents of newborns get the same amount of sleep.
Public records can accurately predict where you live to within a few meters. So can following you home, and asking your friends. I'd be much more "worried" about those things than Foursquare.
I don't get it, unless I misunderstand foursquare....duh? I've never used it but don't you like check into locations? So you probably live near most of the locations you check in to..at least within 50km? Within 50km doesn't seem very impressive.
I don't use it.
Is this the new Facebook? A brief summary would be nice in the ... summary.
I want to delete my account but Slashdot doesn't allow it.
First, I'm not worried in the slightest, since I don't use the useless piece of software.
But knowing where you live isn't the problem. As you say, it's easy to figure out.
No, the problem is suddenly everyone is that guy from that old ADT commercial. They know when you home, and when you not.
And they probably know a guy who can pick most locks with a credit card. Failing that, for a small cut, they can probably get that guy who, in the event of a deadbolt, would just smash the door.
Dude, it's idiot savant, not cosmonaut. Get it right!
Bio questions? Ask me to start a Q&A journal. Computer analogies available for most topics!
"With an accuracy of 50km"
That's not particularly accurate at all - for me, that encompasses parts of seven counties and parts of two major cities (neither of which I live in). In of the metropolitan area of one of the major cities (Seattle), there's probably two or three dozen towns of notably size...
Yeah uh, how is this impressive? I'd be pretty surprised if you couldn't figure out generally where someone lived in from Foursquare, considering for most people, most of their check-ins will be in the city/town in which they live. I mean seriously, 50 km accuracy? My mid-sized (400,000 population) city is around 40 km north to south, and is the only logical place where someone would live in this area (no other significant settlements for at least 100 km in any direction), so that's obvious. And in more rural areas there'd be 50km at least between towns here (Australia), so again it makes it bleedingly obvious where someone would live. In "dense rural" areas common in Europe and North America where there's lots of separate small towns close together this might be a bit more impressive, but still...
I clicked on the link expecting a method whereby they got it down to a particular neighbourhood/a couple of km accuracy.
As an example, checking my own Foursquare profile, out of my total checkins:
- 1 is in Hong Kong
- 2 are in Macau
- 2 are in France
- 3 are in Canada
- 7 are in the UK
- 14 are in Singapore
- 33 are in the United States
- 152 are in Australia (home)
So the home country should be fairly obvious from that. And then of the 152 Australia checkins, 68 are in my home city, which is substantially more than any other single city or town. And that's only looking at checked-in places without considering how OFTEN I check into them. If you look at those figures it becomes even more apparent: the places with the most check-ins are my work and the local airport.
This gets posted on and off to a ton of submissions, it seems, but this may be one of the few where it's actually more interesting than the original post.
If I can paraphrase the article, it would be: "researchers have found that when a site encourages you to publish your GPS coordinates for all of your trivial daily tasks it's not hard to figure out approximately where you live". BRILLIANT!
Well duh. I would hope they could get it a lot closer than that. I mean when I lived in Philadelphia, almost every place I went to was within 3Km of my house. At first I thought they were going to say they *actually* know where you love (i.e. what house), *then* I would have been impressed.
Someone needed an excuse for some grant money to burn methinks.
Science advances one funeral at a time- Max Planck
Between Twitter, Foursquare, and Facebook's timeline; if you can get friended by the object of your ~amour~ (and if they post/update frequently), you practically have a 24-hour electronic watch in-place. You kids have it so easy these days...
Too many people here saying it's obvious and trivial.
Saying it is easy does not make it so. Academic research is often about finding precise quantitative methods to realize intuitive goals by thus explaining and formalizing the original intuition.
Newton "explained that objects fall to the ground": easy? No, because he actually used quantitative models and knew how and to what accuracy he could compute predictions.
Same for this paper.
My book: Friendly F#, fun with game development and XNA; my game: Galaxy Wars by VSTeam; my gamedev language: Casanova.
How about we just do IP geolocation of the top two IP addresses each user logs into ANY given service on the web with. Odds are one is "home" and the other is "work"... 50km is pretty damn large for Foursquare, there has been geolocation research which has gotten it down to within a city block or so.
"Write witty paper about it" ...
#PROFIT!
I, for one, am highly impressed that they only managed 78% with a 50 km margin of error.
That must have taken a real effort to be so inaccurate.
The right to offend is central to the right to free speech.
So Foursquare built-in a google name-search function?
Oh wait, no, this would resolve a person's home with much HIGHER reliability and detail.
-Styopa
Atleast for a small town, it gave me a "moved out of basement" badge
Fuck Beta
What is foursquare? Seriously.
The summary has nothing to do with the actual report.
The data gathered includes mayorships, tips, dones and the home city the user entered, all available from their public API. I'd say using the person's entered hometown is a much better predictor of where they live, especially since 99% of the people entered valid geographical data. Not necessarily correct, but valid in that Yahoo! geolocation could resolve it unambiguously (given exceptions like "Springfield" which is a common enough city name that they ignored it). Of course, that is not on the user's profile page, so right off the bat this is a purely academic exercise.
First, they are trusting the user. Second, the purpose of the study is to evaluate models, not to actually find where users live. Again, academic exercise.
The methodology cannot be the same as the conclusion to qualify as proper research. I could call this team idiots, but it is more likely to again reiterate they are taking an assumption and comparing various models to find out how accurete they are, given the assumptions.
In fact, each model maxed out at 60% accuracy for home city, and it's only by using the best model for each person that they can have a meta-model to identify a location.
And now for the blurb that produced the summary and article:
46% of the results were under 50 km away from the user's reported city, and thus probably correct, especially for the most populous cities which tend to be larger. That means 54% were 50 km or more away. The margin of error is not directional, it is a scalar value not a vector. You could be 50km north or 50km south. I don't believe this results in 50 km^2, this sounds like 100km^2. If it puts you right between two cities 100km apart, there is no way to know which city you call home.
Now, note that they are not finding the user's city, but a city within 50 km of where you actually live. And if you take it as scalar, 100 km^2 of where you actually live.
So the end result of the study is that user-supplied information matches user-supplied information to within 100km^2 78% of the time. Which is piss-poor.